The Construction of a Mountain Vegetation Knowledge Graph Incorporating With Geographical Principles, Maps, and Remote Sensing Images

A great deal of geoscience knowledge exists in the form of unstructured text or maps, which are difficult to use by structured models or to process by computers. Thus, it is urgent to transform them to structured knowledge graph (KG). However, the development of geoscience KG (GKG) lags behind the general KG because it involves in the complexity of spatiotemporal relationships and knowledge from multisources. This study constructed a mountain vegetation KG (MVKG) incorporating with vegetation geographical principles, maps, and remote sensing (RS) images with the support of ArcGIS and deep learning method to facilitate the use of vegetation knowledge in various disciplines. The results showed that: 1) for the construction of a GKG, such as the MVKG, it is first necessary to define a strict and compatible ontology to classify and organize all the knowledge in order to facilitate structured representation and storage of them; 2) the MVKG entities were labeled from vegetation maps with the support of ArcGIS, which indicated that the spatiotemporal representation, organization, and analysis techniques of GIS can effectively support the construction of the GKG; 3) the RS image features extracted by the deep learning method were embedded into the properties of the MVKG entities, which will be significant for the MVKG application because RS monitoring is indispensable for the study of vegetation distribution and changes. The MVKG can also enhance the application of vegetation knowledge and information in RS monitoring for vegetation cover and change, mountain ecology, and climate change.